In this paper, we introduce a divide-and-conquer algorithm to improve sentence alignment speed. We utilize external bilingual sentence embeddings to find accurate hard delimiters for the parallel texts to be aligned. We use Monte Carlo simulation to show experimentally that using this divide-and-conquer algorithm, we can turn any quadratic time complexity sentence alignment algorithm into an algorithm with average time complexity of O(NlogN). On a standard OCR-generated dataset, our method improves the Bleualign baseline by 3 F1 points. Besides, when computational resources are restricted, our algorithm is faster than Vecalign in practice.